ARIS-Anything

Research as 科研 (kēyán) covers academic science.
Research as 研究 (yánjiū) covers everything else — investment, law, product, market, learning, investigation, journalism, engineering post-mortems, ...

ARIS-Anything explores how the ARIS methodology — cross-model adversarial review, skill-based workflows, persistent knowledge wikis — generalizes from academic research to any structured inquiry.

📖 中文版 (Chinese version): README_CN.md

🏆 Why this isn't vaporware

ARIS-Anything is an exploration, not a from-scratch venture. The methodology is already battle-tested at scale:

· · · · ·

• ⭐ ~10k GitHub stars on the main ARIS repo — top-trending AI research agent
• 🥇 HuggingFace Daily Papers #1 — arXiv:2605.03042 top of the day
• 🏆 AI Digital Crew · Project of the Day (2026-03-14)
• 📰 Featured on PaperWeekly + awesome-agent-skills
• 🛠 75+ research skills spanning idea exploration → experiments → papers → rebuttals → talks
• 🌐 7+ platforms supported — Claude Code · Codex CLI · Cursor · Trae · Antigravity · GitHub Copilot CLI · OpenClaw
• 🔧 ARIS-Code standalone CLI — multi-provider runtime, no Claude Code dependency required
• 📚 First downstream repo already shipped two artifacts: ARIS-in-AI-Offer bundles two production deliverables on the same /render-html workflow — (1) 23 bilingual interview cheat sheets (the original), and (2) ARIS-Homepage v1 (/homepage-generator skill: CV → fact-checked single-file academic homepage; DBLP/arXiv hard-audit on venue/year/author mismatches and fabricated awards). Two completely different deliverables, one workflow — proof the ARIS pattern carries beyond science papers.

ARIS-Anything is the next reach: take the proven loop, point it at non-academic structured inquiry. Everything below is grounded in what already works — not a wishlist.

🌟 The Premise

ARIS — Auto Research in Sleep was built for academic research: find ideas → run experiments → write papers → handle rebuttals. But the underlying loop is general:

1. Plan the inquiry — structure the question, scope, deliverable
2. Draft the work — executor model produces the first version
3. Review adversarially — a different-family reviewer model finds flaws (Claude × GPT-5.5 xhigh × Gemini)
4. Iterate until convergence — fix → re-review → ship
5. Persist — write findings to a versioned wiki so future-you starts from the cumulative state, not zero

The Chinese wordplay matters: 科研 ⊂ 研究. Academic research is one specific shape of inquiry; the broader 研究 covers any structured investigation where you have a question, evidence, and a deliverable. The same five-step loop carries over.

🚀 Incoming siblings

The "Anything" exploration runs in parallel with several domain-specific repos under early development. Each one re-uses the ARIS core loop with a domain-appropriate deliverable shape — same five-step inquiry, different cover sheet:

These are scheduled rather than promised — code drops in this repo and the matching wanshuiyin/ARIS-* repos when ready. If one of these domains is your day job, open an issue — you can shape the API.

🧭 Where the loop applies

Same loop, different cover sheet:

The 75+ skills in ARIS — /research-lit, /idea-creator, /proof-checker, /research-wiki, /peer-review, /citation-audit, /kill-argument, ... — were built for science, but most of the structural moves generalize. This repo is where we collect the adaptations.

🧱 Core invariants (preserved from main ARIS)

These do not change across domains:

• Cross-model adversarial review — executor and reviewer must come from different model families. No LLM ever judges its own output.
• Reviewer independence — only the artifact is passed to the reviewer, never the executor's summary or fix list. Fresh thread per round.
• Persistent wiki — accumulated findings live in a typed knowledge wiki, not in conversation context. Future runs start from the cumulative state.
• Skill-based composition — every workflow is a SKILL.md contract. Re-use, fork, and remix freely.

What changes per domain: the deliverable shape (a paper vs. an investment memo vs. a legal brief vs. a study plan), the evidence sources, and the claim taxonomy.

🌱 Status

🌿 Early genesis (2026-05). This repo is a placeholder for an exploration: which ARIS skills generalize, which need a re-skin, which need fresh design.

If you have a domain in mind:

1. Pick the closest existing ARIS skill from the table above
2. Try it on your domain — note what breaks, what fits
3. Open an issue with the friction points
4. We'll figure out together what the "Anything" version should look like

Early focus areas being prototyped:

• Investment due diligence — equity research with cross-model review of bull/bear claims
• Legal case analysis — argument-structure auditing via /proof-checker on legal logic
• Self-directed learning — re-using /interview-cheatsheet for arbitrary curricula
• Personal post-mortem — applying paper-review skills to incident reports

🔗 Related repos

💬 Community

Shared community with the main ARIS repo — same WeChat group covers methodology discussions, the AI-Offer cheat sheets, and this exploration:

License

MIT — open, fork-friendly, remix-friendly. The methodology is the gift; the implementations should be everyone's.